Run model

After the installation, the command line script pet_run is available, and can be used to run fitted PET models on the new (test) structures. The usage is the following:

$ pet_run <structures_path> <path_to_calc_folder> <checkpoint> <n_aug> <default_hypers_path> <batch_size> --path_save_predictions=<your_path>

Structures should be formatted in the same way as training and validation ones, as discussed in the “train model” section. The Calc folder is a path to a folder with checkpoints which is created by the “train_model.py” script. <checkpoint> refers to a specific checkpoint to be used. PET saves several checkpoints, such as the one with the best MAE in energies on validation or the best RMSE in forces on validation, which can happen on distinct epochs. Run python3 estimate_error.py --help to see the full list. <n_aug> is a number of rotational augmentations during inference. Finally, one can optionally specify the path where predicted energies and forces are to be saved as numpy (.npy) arrays.